Andrew Gordon Wilson

Joining our Group

It is safe to assume I am taking PhD students and postdocs, unless I explicitly specify otherwise on this page. At the same time, admissions into our group is exceptionally competitive. The best way to improve your chances of admission is to make a detailed genuine case in your formal application about why our group, and me as an advisor, are the best fit for your interests and plans, and how you expect to contribute to the group. In order to make that case, you will need to meaningfully engage with the work we are doing, and understand our vision.

Fundamentally, we aim to develop a prescriptive approach to building autonomous intelligent systems. This effort involves a variety of different research initiatives, which cumulatively work together towards achieving this vision. A major theme that unifies many of these initiatives is progress towards an actionable understanding, so that we can select for particular properties aligned with human goals, including safe and reliable decision making. These areas, and some example papers, include:
Understanding deep learning models, including LLMs and vision models
[e.g., 1, 2, 3, 4, 29, 30, 31]
Uncertainty representation, Bayesian methods, online decision making
[e.g., 1, 5, 6, 7]
Distribution shifts, spurious correlations [e.g., 8, 9, 10, 11]
Encoding and learning inductive biases (e.g., equivariances) [e.g., 12, 13, 14, 15]
Numerical linear algebra for scalable inference [e.g., 16, 17, 18, 19, 20]
Machine learning for physics, and physics for ML [e.g., 21, 22, 13, 20, 15]
Simple practical methods [e.g., 23, 24, 25, 26, 4]
Scientific discovery (protein engineering, materials design) [e.g., 27, 28, 32]

If you wish to apply, I recommend reading some of our papers carefully, and describing how your interests connect to our work in your application. A full list of papers is available at my Google Scholar. I also recommend the following video lectures, for a sense of my approach, and an overview of some of our work. While many topics are covered, much of our work, for example on equivariance modelling, LLMs, and distribution shifts, is not described in these videos:
(1) My introduction to Bayesian machine learning, which opens with background on how I started and how my interests have evolved.
(2) ICML Bayesian Deep Learning Tutorial.
(3) Lecture on Representation Learning with Gaussian Processes.

I advise students in Courant Computer Science (Dec 12 deadline), Mathematics (Dec 18 deadline), and the Center for Data Science (Dec 6 deadline).

For Fall 2024 admission, I will primarily be considering applications through Courant Computer Science. If you apply to math, feel welcome to let me know you applied. In general, you are free to e-mail me, but please only do so if you have carefully read some of my papers and believe there is a particularly compelling and specific connection with my group. I will not be able to reply to any generic messages. In general, do not be discouraged by a lack of response, as I receive many more e-mails than I can respond to. It is most important to simply list my name in your formal application.

A subset of initiatives I am particularly excited about this year include AI alignment (especially weak-to-strong generalization, truthfulness, and disentangled representations), numerical methods as a foundation for ML (see the CoLA project), understanding and generalizing large language models, and machine learning for science.

Postdoctoral Fellow Applicants: E-mail me your CV and a short description of how your interests and plans connect to my group. The best way to work with me as a postdoc is through the CDS faculty fellowship program. But be mindful of the deadlines! The Courant CS faculty fellowship is also a nice option. I'd recommend applying to both.

Undergraduates: If you're at NYU, and you have done courses in machine learning, probability, and linear algebra, you can send me an e-mail with your CV.